Active and weakening MJO events in the Maritime Continent

Abstract

To better understand the Madden–Julian Oscillation (MJO) in the Maritime Continent (MC), events from 1980 to 2019 were classified as active or weakening according to propagation characteristics and amplitude changes in two leading indices, the Real-time Multivariate MJO (RMM) index and Outgoing Longwave Radiation (OLR) MJO Index (OMI). Active MJO events had larger index amplitudes than weakening events, and those amplitude differences appeared at day − 3 in the RMM and day − 8 in the OMI and remained through day + 15 in both indices (day 0 marked the day an event entered the MC). Thus, active events were stronger than weakening events as they approached and crossed the MC. To understand differences in environments between these events, composites of OLR and specific humidity were compared for each. Active MJO events had consistently more negative OLR anomalies and greater specific humidity in both the boundary layer and free troposphere than weakening MJO events. Those differences persisted during both easterly and westerly phases of the Quasi-biennial Oscillation (QBO) and during La Niña and El Niño. Active MJO events during QBO westerly winds had more specific humidity than active events did during QBO easterly winds. Finally, both active and weakening MJO events featured a diurnal cycle of lower-tropospheric specific humidity over land areas of the MC, and the amount of atmospheric moisture available to active events was consistently greater than weakening ones. These results indicate that index amplitude, OLR, and tropospheric specific humidity can all be important tools in identifying active and weakening MJO events in the MC, including for different phases of the QBO and the El Niño-Southern Oscillation (ENSO).

Interpretable Deep Learning for Probabilistic MJO Prediction

“…(2015) and Barrett et al. (2021) found OLR precursors in this region which distinguished between propagating and non‐propagating MJO events. We conclude that the CNN has identified true predictive features of MJO propagation, giving us confidence in the network.…”

“…The RMM1 signal anomaly is greater than for SHUM400, and it is stronger over the Pacific than was the case for SHUM400. Both Feng et al (2015) and Barrett et al (2021) found OLR precursors in this region which distinguished between propagating and non-propagating MJO events. We conclude that the CNN has identified true predictive features of MJO propagation, giving us confidence in the network.…”

“…As inputs we use subseasonal anomalies of OLR, UA200, and UA850, consistent with fields used to compute the RMM indices. We supplement these with four further fields which provide complementary information: daily mean Specific Humidity at 400 hPa (SHUM400) was included because Barrett et al (2021) reported large differences in SHUM400 between MJO events which propagate and weaken over the Maritime Continent; daily mean geopotential at 850 hPa (Z850) provided skill in previous work (Toms et al, 2020); daily mean downwelling longwave radiation at the surface has a marked annual cycle, which we found a more effective means of accounting for the seasonality of the MJO than including a dummy variable. Finally, daily anomalies of sea surface temperature (SST) are included, since the MJO is known to be linked to El Nino-Southern Oscillation (ENSO: e.g., Kessler, 2001).…”

“…If the probabilistic forecast mean is assessed, averaging out the unpredictable “noise,” current dynamical models have a prediction skill up to 3 weeks (Lim et al., 2018; Vitart, 2017). However, systematic biases remain, especially in the propagation of the MJO convective anomaly over the Maritime Continent (Barrett et al., 2021; Kim et al., 2016; Li et al., 2020). In contrast to the mean skill, the probabilistic skill of MJO forecasts is low (Lim et al., 2018; Vitart, 2017).…”

Estimates of the sensible heat of rainfall in the tropics from reanalysis and observations

“…An example of Q P and Q SH during the passage of a Madden‐Julian Oscillation (MJO, Madden and Julian, 1971) event in April 2009 (Figure 6) across the Maritime Continent region is shown in Figure 7. This spring MJO event (Tan et al ., 2020, 2021) represents a typical event (Barrett et al ., 2021) that propagated close to the equator and produced significant precipitation in the MC region, exceeding 15 mm day −1 over some parts of the Maritime Continent. The area also provides an opportunity to describe the magnitude of Q P over both land and oceanic areas including the coastal land and coastal oceans.…”